Skill Creation
Scenario
You’ve just completed a complex task — say, generating a competitor analysis report. It took a few iterations to get the format and data sources right. Now you want to save this workflow so you (or your team) can rerun it with a single command, without re-explaining everything.Materials
- A completed conversation where the agent produced a satisfactory result
- Clarity on what parts should be parameterized (e.g., competitor name, date range)
Steps
1
Complete a Task Successfully
Work with the agent until you have a result you’re happy with. The conversation history becomes the raw material for the Skill.
2
Ask the Agent to Create a Skill
In the same conversation, say:
“Turn this into a Skill. Name it ‘Competitor Analysis Report’. The user should provide the competitor name and date range as inputs.”The agent analyzes the conversation, extracts the workflow logic, and generates a Skill definition.
3
Review and Edit
The Skill appears in Settings → Skills with the extracted prompt, parameters, and description. You can edit any field directly in the GUI.
4
Use Your Skill
Trigger it in any conversation with a slash command:
/competitor-analysis-report
Or select it from the Skill panel. The agent loads the Skill’s instructions and asks for the required inputs.Skill Evolution: Over time, the agent can automatically refine Skills based on feedback and usage patterns. See Skill Evolution for details.
Deliverable
A reusable Skill stored in your agent’s Skill library, callable via slash command or GUI selection.Acceptance Criteria
- Skill produces results consistent with the original conversation
- Input parameters are correctly extracted and prompted
- Skill is visible in Settings → Skills
- Slash command activates the Skill correctly
Package a workflow as a skill pack (ad creative example)
A skill pack is a small set of focused skills that hand off inputs and outputs. It is the right shape when a workflow has fixed stages: each skill stays short, and the agent loads only the stage it needs instead of one giant prompt. Example — a 5-stage ad creative pipeline:
Build it in five steps:
- Create a folder
ad-creative/with oneSKILL.mdper stage; keep each contract narrow (input, steps, output file, success criteria, traps). - Put shared state in one workspace file such as
ad-context.md— brand voice, audience facts, past winners. Every stage reads and updates it, so context compounds across runs without stuffing every prompt. - Adopt the folder via Settings → Skills → Local skill paths, or describe the workflow to the agent and use
/learnto draft the skills. - Review drafts in Pending Evolutions; approve, edit, or reject each one.
- Run a small A/B in Eval Lab before trusting the pack, then bind a Weekly Sprint cron if you want a recurring cadence.
Myrm does not ship a dedicated ad-creative pack today. You can compose the existing Social Media Strategist, Writer, and Designer agents, or author the five skills yourself with the steps above. Treat any “comment MAP to get a PDF” pack as content, not a tested capability.
What’s Next
- Discover community skills → Skill Discovery & Registry Mirror
- Bind to a scheduled task → Combine with Scheduled Automation to run the Skill on a recurring schedule